The work that would bring signups keeps losing to your week.

A team of AI agents that works toward a marketing goal you set, such as “40 signups a month from content”. Kovnor proposes the agents, each with one job, a lead that owns the plan, and a reviewer that checks work before you see it. A practice run shows drafts, sources and cost before anything is sent. After that the team keeps going, and you’re asked only when a person is needed.

Once the team is set up, work starts on its own at each check, day and night. The lead agent hands out the work and changes the plan when the numbers say so, a reviewer checks the work, and Kovnor reassigns work that gets stuck. You hear from it when it needs a person, and Home shows what got done.

The standing jobs that eat your week: posts and pages that bring signups, a morning sweep of mentions, competitors and threads worth joining, and replies in your voice. Each is a project with an outcome Kovnor checks, a lead agent that owns the plan, and a reviewer that checks the work before you see it.

Before anything goes out, your new team works on public data and hands in about ten drafts, each with its sources and what it cost. You accept, edit or reject them. Once three are accepted you can go live; until then nothing is sent.

Only when it takes a person: an approval, a file, a login, a decision about money or something new. Each ask has a recommendation and a default with a deadline, so nothing stalls if you’re busy. Hand any ask to a teammate, or answer it from your phone.

Every agent starts by asking. When its work is approved without changes again and again (for example 20 replies in a row), Kovnor offers to let it do that one thing on its own. Sending anything new, spending over your limit and deleting still ask, and you can undo it any time.

Each of those is one assistant you ask for tasks. Kovnor starts from an outcome and proposes a team: each agent with one job, a lead agent that owns the plan, and a reviewer that checks the work before you see it. Every result shows its cost, and your limits pause the work instead of billing past them. The side-by-side pages are under Compare.

Not yet, and it will stay optional. Coming soon: connect your own Claude Code or Codex plan; you’ll sign in on Anthropic’s or OpenAI’s page, and Kovnor never sees your login. Today, add an Anthropic or OpenAI API key (your provider bills the model usage, so credits go much further) or run on Kovnor credits. Any MCP client can already connect to Kovnor as one of your agents: Claude Code and Codex with one command, Cursor and others by adding one URL. You sign in to Kovnor and approve it.

Three monthly plans: Build $99 with $100 of credits, Grow $399 with $450 and Scale $699 with $850. Credits pay for coordination, computers, and model usage when it runs on Kovnor; top up any time or turn on auto top-up. Every plan has unlimited team members. Model usage on your own API key is billed by your provider, and limits are always honoured.

Every result shows what it cost, by agent and project, and a cost nobody recorded says “unknown”, never $0. Set limits for the workspace, each project and each agent; at a limit the work waits and asks you. Pause everything is one button on Home: running work stops at a safe point, nothing sends, and nothing is lost.

No. The product pictures are illustrations with example names and numbers, labelled as such. We’ll publish real numbers when we have measured them.

Book a demo. On a 20-minute call with the founder we set up your first goal, walk through the practice run, and turn the team on when you approve it. Plans and limits are agreed on that call before anything goes live. Or create a workspace in the app: 7-day free trial on monthly Build, Grow, and Scale: pick a plan at the end of onboarding, add a card at checkout, and you are billed when the trial ends unless you cancel. One trial per person; yearly billing has no trial.